feat(daily-fact): add reasoning_depth A/B experiment + fix compiler expression transform for inline mode

[Copilot]

Implements the reasoning_depth experiment on daily-fact to test whether a multi-candidate deliberation step produces more novel/engaging verses than the current single-pass approach. Also fixes a latent compiler bug exposed by this workflow's inlined-imports: true setting.

Changes

daily-fact.md — experiment wiring

• Added experiments.reasoning_depth frontmatter block (single_pass vs multi_candidate, 50/50, starts 2026-05-11, issue [ab-advisor] Experiment campaign for daily-fact: A/B test reasoning_depth #31324)
• Replaced the single Favor recent updates guideline bullet with a variant-conditional block:

{{#if experiments.reasoning_depth == "multi_candidate"}}
- **Multi-candidate deliberation**: Before writing, identify exactly **3 distinct candidate facts** ...
  score each 1–5 on novelty and poetic potential, write verse for the highest-scoring one only.
{{else}}
- **Favor recent updates** but include variety - pick something interesting, not just the most recent
{{/if}}

daily-fact.lock.yml — regenerated via gh aw compile daily-fact

Compiler bug fix: expression_extraction.go

transformExperimentsExpression only handled bare ${{ experiments.NAME }}. In workflows with inlined-imports: true the compiler pre-processes all ${{ … }} expressions before runtime; the comparison form ${{ experiments.NAME == "value" }} was left untransformed, producing an invalid GitHub Actions expression where experiments is not a real context — always evaluating to false, silently killing the treatment arm.

Fix: added experimentComparisonRegex to rewrite all four comparison operators (==, ===, !=, !==):

experiments.NAME == "value"  →  steps.pick-experiment.outputs.NAME == "value"

The generated env var now produces a valid GA expression:

# Before (always false — experiments context doesn't exist in GA)
GH_AW_EXPR_70B2292B: ${{ experiments.reasoning_depth == "multi_candidate" }}

# After (evaluates correctly from pick-experiment step output)
GH_AW_EXPR_DD64DF46: ${{ steps.pick-experiment.outputs.reasoning_depth == "multi_candidate" }}

Four new test cases cover ==, ===, !=, !== in expression_extraction_test.go. This bug would have affected any workflow combining inlined-imports: true with value-comparison experiment conditionals.

[Copilot]

Pull request overview

Adds an A/B experiment (reasoning_depth) to the daily-fact workflow and fixes inline-mode experiment expression rewriting so variant conditionals evaluate correctly under inlined-imports: true.

Changes:

• Introduce experiments.reasoning_depth (50/50 single_pass vs multi_candidate) and gate a new “multi-candidate deliberation” guideline via a variant-conditional block in daily-fact.md.
• Update the compiler’s transformExperimentsExpression to also rewrite experiments.<name> {==,===,!=,!==} "value" comparisons to steps.pick-experiment.outputs.<name> ....
• Regenerate daily-fact.lock.yml and add tests covering the comparison-operator rewrite cases.

Show a summary per file

File	Description
pkg/workflow/expression_extraction.go	Extend experiment expression rewriting to handle string comparisons in inline mode.
pkg/workflow/expression_extraction_test.go	Add unit tests for ==, ===, !=, !== experiment comparison transforms.
.github/workflows/daily-fact.md	Wire reasoning_depth experiment and add variant-conditional prompt guidance.
.github/workflows/daily-fact.lock.yml	Regenerated compiled workflow reflecting experiment wiring and fixed expression transform.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 4/4 changed files
• Comments generated: 1

[pelikhan]

@copilot review all comments

[Copilot]

@copilot review all comments

Fixed in the latest commit — switched the nested double-quoted example on line 220 to use a backtick so the rendered comment is unambiguous: `experiments.<name> == "value"`.